Packaging Model using Registry Environment fails during ACR pull

[jakeatmsft]

Operating System

Windows

Version Information

azure-ai-ml -1.12

Steps to reproduce

1. Build environment in shared registry
2. Create a model package to use shared environment as base environment

• Package.yml :
• $schema: http://azureml/sdk-2-0/ModelVersionPackage.json
  base_environment_source:
  type: environment_asset
  resource_id: azureml://registries/shared-registry/environments/shared_base_env/versions/1
  target_environment_name: model-pkg
  inferencing_server:
  type: azureml_online
  code_configuration:
  code: model_root
  scoring_script: score.py

3. build model package in workspace.

Expected behavior

New model package env created in local aml workspace successfully

Actual behavior

NOTE: Azure Container Registry path has domain printed twice.

Error in Step 1 of environment build:

Step 1/6 : FROM 4*******.azurecr.io/ 4*********.azurecr.io**/azureml/544/shared_base_env:1@sha256:7824

Get https://4*********.azurecr.io/v2/4*********.azurecr.io/azureml/54*********4/shared_base_env/manifests/sha256:7824*********
: unauthorized: authentication required

Addition information

No response

[santiagxf]

Hi @jakeatmsft! Apologies for the delay in the response! What's the type of the model you are working on?

Looks like this model is of type custom since you are indicating an scoring script and you are probably working on a workspace with private link. We have a known issue about packages returning an access denied error when trying to reach ACR. We are working to solve this issue soon. The workaround to this is to enable annonymous pull access at ACR as explained here: Enable anonymous pull access

az acr update --name myregistry --anonymous-pull-enabled

We hope we can fix this limitation soon.